Group Signature Federated Learning for Anonymous Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Federated learning systems face increasing computational and communication costs as the number of clients increases, and there is a risk of personal information exposure due to client tracking during decentralized learning.
Innovation Solution
A group signature-based federated learning method and system where clients request group signatures from an administrator terminal, generate update information using these signatures, and a server verifies and redistributes the federated learning model based on matching signatures, reducing computational and communication costs while protecting personal information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a decentralized learning structure is used to avoid single-point failure, then system reliability is improved, but computational and communication costs increase exponentially
Solution Approach 1:
The patent introduces a central server as an intermediary that coordinates the decentralized learning process. The server aggregates model updates from multiple clients and manages the federated learning workflow, allowing the system to maintain decentralization benefits while avoiding exponential cost increases through centralized coordination.
Solution Approach 2:
The learning process is segmented into distinct phases: local model training on client devices, update aggregation on the server, and global model distribution. This segmentation allows computational work to be distributed across clients while the server handles only the aggregation step, preventing exponential computational cost growth.
2Loss of information
If each client communicates with multiple clients to maintain confidentiality, then information security is improved, but communication costs increase exponentially
Solution Approach 1:
The central server acts as an intermediary that receives encrypted model updates from clients and aggregates them without requiring direct client-to-client communication. This eliminates the need for each client to communicate with multiple other clients while maintaining security through encryption and secure aggregation protocols.
Solution Approach 2:
The communication burden is extracted from the client-to-client interaction model and consolidated into client-to-server communication. Each client communicates only with the central server, reducing communication complexity from exponential to linear scaling while the server performs the aggregation function that would otherwise require extensive peer-to-peer communication.
3Loss of information
If clients exchange encryption keys directly, then confidentiality is maintained, but user tracking becomes possible
Solution Approach 1:
The central server mediates the key exchange process by distributing encryption keys to clients without requiring direct key exchange between clients. This maintains confidentiality through secure key management while preventing tracking, as the server uses anonymous authentication mechanisms that do not link clients to their specific communication patterns.
Solution Approach 2:
The central server provides multiple functions: it acts as a key distribution center, an aggregation point for model updates, and an anonymous authentication service. By consolidating these functions, the system maintains confidentiality through secure key management while preserving client anonymity through unified, non-traceable authentication mechanisms.
Data Source
AI summary
According to one embodiment of the present invention, a group signature-based federated learning method by a group signature-based federated learning system including at least one client terminal and an administrator terminal provided as a group, and a server distributing a federated learning model to the at least one client terminal may comprise requesting, by the client terminal, a group signature to the administrator terminal; transmitting, by the administrator terminal, a group signature response to the client terminal in response to the request for the group signature; generating, by the client terminal, update information of signed federated learning using the group signature response and transmitting it to the server to request participation in federated learning update; transmitting, when the server receives the request for participation in the federated learning update, a latest group signature request to the administrator terminal and receiving a latest group signature response from the administrator terminal; and comparing, by the server, the latest group signature response with update information of the signed federated learning to determine whether signatures match, and if the signatures match, updating and redistributing the federated learning model to at least one client terminal.